Spaces:
Sleeping
Sleeping
Commit History
-
Add missing sentencepiece dependency and improve tokenizer loading
lowvoltagenation +1: Claude- Added sentencepiece>=0.1.99 to requirements.txt (required for Mistral tokenizer) - Added protobuf and tiktoken dependencies for better tokenizer support - Improved tokenizer loading with fast/slow fallback logic - Maya LoRA should now load properly with working tokenizer The model was loading (checkpoint shards 100%) but tokenizer was failing. This should resolve the sentencepiece tokenizer error. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>1c69287 -
Restore Maya LoRA for proper gated model access with Pro account
lowvoltagenation +1: Claude- Revert Maya LoRA to inference_api type with requires_auth: True - Use original Mistral v0.3 base model as intended - User has HuggingFace Pro account and should be able to access gated models - Need to configure HF_TOKEN in Spaces secrets for authentication Next step: Add HF_TOKEN secret in HuggingFace Spaces settings. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>45bd7a8 -
Add LoRA fallback system for Maya model compatibility
lowvoltagenation +1: Claude- Switch Maya LoRA from inference_api to local LoRA loading - Add fallback base model system (v0.1 first, v0.3 if needed) - Add _load_as_inference_api() fallback method if LoRA fails - This should handle the gated v0.3 issue by trying non-gated v0.1 first - Provides multiple fallback paths for Maya LoRA loading The system will try: 1. Load LoRA with non-gated Mistral v0.1 base 2. Fallback to gated Mistral v0.3 base if needed 3. Fallback to inference API if LoRA loading fails entirely 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>cbc3ed8 -
Fix Maya LoRA to use non-gated Mistral v0.1 base model
lowvoltagenation +1: Claude- Updated Maya LoRA base_model reference from v0.3 to v0.1 - Mistral v0.3 became gated and requires approval - Mistral v0.1 is non-gated and should work without authentication - Maya LoRA should now load properly via inference API 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>6ce1953 -
Restore Maya LoRA model: Use inference API to avoid base model download issues
lowvoltagenation358886d -
Fix gated model issue: Use Mistral v0.1 with Maya personality via system prompt
lowvoltagenationcb29bc9 -
Fix HF Spaces auth: Use inference API for Maya model to avoid gated base model download
lowvoltagenationfca6913 -
Fix Maya LoRA responses: simplify prompts, add proper formatting, improve cleaning
lowvoltagenation6895a87 -
Add support for LoRA model loading in ModelInterface
lowvoltagenation- Updated requirements.txt to include 'peft' library. - Enhanced ModelInterface to load LoRA adapters with base models, including error handling and tokenizer setup. - Integrated logging for model loading processes to improve feedback during operations.2043924 -
Update model dropdown to show only Maya LoRA, Mistral v0.3, and Kimi K2
lowvoltagenationed31be4
-
adding maya-gradio files
ba20783